Bharat manages IT operations at Freewheel. He is in charge of two teams: one focuses on enterprise applications while the other on vendor management, IT procurement, and FinOps. He has been doing AWS cost management since 2017. This was initially unstructured but is becoming more structured with regards to budgeting.
His forecasting journey started out using trend-based forecasting utilizing statistical formulas without engaging engineers. However in the March 2021 time frame his team saw large discrepancies due to traffic changes related to business reasons. His team decided to change to a bottoms up approach using a monthly survey to the account owners to validate the automatically generated forecasts.
While behind the scene forecasting produced reasonable results, it was not able to handle special business events. Using business drivers to correlate revenue to increases in AWS cost, we were able to adjust cloud budgets. We worked with the CTO office to get support and then reached out to the ~90 account owners to meet every two weeks to review forecasts. Attendance is high, and we focus on the main drivers. The account owners are aware of the forecasting formulas we use. We achieve 82% to 85% forecast accuracy, which is sufficient for Finance.
Our learnings are the following: It really helps to see trends and map events to specific months. You need good training around tagging e.g. when hiring engineers they need to know about tagging and how important it is. Culture is hard to change, you need to set expectations at the beginning. Automation can be retro fitted, for example right-sizing EC2 instead of buying more savings plans.